Researchers have taught machines to interpret scents with growing precision. Now one system aims to sort out why a patient gasps for air the moment they enter the emergency department.
Ainos, a company listed on NASDAQ under the ticker AIMD, has teamed up with National Taiwan University for a year-long study launching in July. The project will collect breath samples from people suffering dyspnea, the medical term for shortness of breath that ranks among the most frequent reasons patients crowd emergency rooms. The goal sits squarely on distinguishing between acute exacerbation of chronic obstructive pulmonary disease and acute decompensated heart failure. Those two conditions demand opposite treatments. Get it wrong and outcomes worsen fast.
The technology at the center of this effort rests on an AI Nose module packed with multiple micro-electro-mechanical system sensors paired to a digital processor. Gases trigger changes in sensor resistance. Those shifts turn into digital signals. A proprietary Smell Language Model then steps in. It learns patterns, classifies them, and places them in context much like the human olfactory system processes odors. The company calls the output Smell ID, a machine-readable label for complex scent signatures.
“AI Nose was originally developed with medical diagnostic applications in mind, where non-invasive sensing, accuracy, and real-world validation are essential,” Ainos CEO Eddy Tsai said in a report from The Register. “This research program brings that experience back into a high-value clinical setting and extends our Smell AI platform into digital breath intelligence.”
The partnership builds on an earlier deployment already running inside an active emergency department at National Taiwan University Hospital. There the same hardware monitors for respiratory infections and helps track overcrowding across waiting areas, treatment rooms, and observation zones. Success in the new dyspnea trial could produce a breathprint database. That resource might later support outpatient clinics or even devices for patients to use at home.
But Ainos does not operate alone in this space. The Wall Street Journal examined the wider rise of electronic noses last March. Its reporting highlighted how these systems scan breath for signs of lung cancer, urinary-tract infections, and gastrointestinal disease with promising consistency. One researcher quoted in that piece noted the sensors can detect aromas with roughly 1,000 times the precision of the human nose by analyzing volatile organic compounds and their combinations. Yet challenges remain. Humidity, the way molecules disperse, and the exact manner of inhalation all complicate readings, warned Haritosh Patel of Harvard.
Similar work appears across the globe. A study published in Analytical Chemistry last year achieved 94.4 percent accuracy distinguishing Parkinson’s disease patients from controls by training an AI model on volatile organic compounds extracted from ear canal secretions. Four specific compounds differed markedly between the groups. The authors, led by Hao Dong and colleagues, suggested the method could eventually become a quick bedside screening tool, according to coverage in Parkinson’s News Today.
Other efforts target ovarian cancer through blood volatiles or use sensor arrays to flag early metabolic shifts. RealNose.ai, an outfit focused on machine olfaction, lists prostate cancer detection from urine scent among its ambitions and has filed patents on receptor selection and saliency-guided optimization. A separate arXiv paper from late 2025 reviewed how stabilized mammalian olfactory receptors combined with neural networks show increasing promise for identifying disease markers in urine or chemical threats in the field.
Ainos itself has pushed the technology beyond hospitals. The company announced deployments of 1,400 AI Nose units in semiconductor backend processes and another 200 under validation in frontend operations. It signed a distribution agreement with Solomon Technology to combine its Smell Language Model with visual systems for robotics and smart manufacturing across Asia. Pilot programs were slated for the second half of 2025 with commercial rollouts targeted for this year. CEO Tsai has described smell as a fresh category in industrial sensing, one that can spot impurities threatening multimillion-dollar chip batches or flag safety issues before human noses notice.
The sensor data feeds into a model trained on a scent dataset gathered over more than a decade. It detects volatile organic compounds down to parts-per-billion levels. Those readings become structured Smell IDs that other AI systems can index, search, or combine with vision and language models. The approach echoes how large language models transformed text. Here the input is chemical rather than linguistic.
Yet the field stays young. Engineers still wrestle with variability across patients, environments, and even the same person at different times of day. Breath analysis must contend with diet, medications, and recent smoking. Earwax or urine samples introduce their own collection and standardization headaches. Most studies remain small. They need multi-center trials across diverse populations before regulators greenlight widespread clinical use.
Still the momentum builds. A dataset project called oMNIST seeks to give olfactory AI the same foundational training resource that ImageNet once supplied to computer vision. Researchers at MIT and elsewhere have assembled libraries of real-world smells captured by portable gas sensors. Their work already classifies food substances or allergens with growing speed.
In emergency medicine the payoff could arrive sooner. Doctors facing a breathless patient often order chest X-rays, blood tests, and echocardiograms that take time and resources. A portable breath analyzer delivering a rapid probability score for COPD flare versus heart failure might speed triage, reduce unnecessary imaging, and start the right therapy minutes earlier. Those minutes matter when lungs or the heart teeter on the edge.
Ainos frames its NTU collaboration as a return to its medical roots after industrial successes. If the breathprints prove reliable, the company envisions expanding the same platform to other emergency conditions, chronic disease monitoring at home, and even infection control in crowded wards. The hardware is compact enough for bedside carts or handheld units. Cloud analytics handle the heavy model lifting.
Critics point out that dogs have performed similar feats for years with little more than their natural noses and some training. Medical detection dogs alert to cancers, low blood sugar, and oncoming seizures. The machines simply aim to scale that capability, standardize it, and remove the handler’s subjective read. Success would not replace physicians. It would hand them one more objective data point in a high-stakes environment.
The Taiwan research begins next month. Results will not appear overnight. A full year of patient recruitment, sample analysis, model refinement, and validation lies ahead. Even positive findings will require replication. But the direction feels clear. Smell, long the neglected sibling of sight and sound in artificial intelligence, has entered the diagnostic conversation.
Companies, universities, and hospitals now race to build the datasets, refine the sensors, and prove the clinical value. The ones who get it right could shift how doctors assess the breathless, the early-stage cancer patient, or the factory worker exposed to unseen vapors. For now the AI nose waits in the lab and the emergency bay. Soon it may start telling doctors what the patient cannot.
WebProNews is an iEntry Publication